Systems and methods for machine learning based physiological motion measurement

Abstract

A system for physiological motion measurement is provided. The system may acquire a reference image corresponding to a reference motion phase of an ROI and a target image of the ROI corresponding to a target motion phase, wherein the reference motion phase may be different from the target motion phase. The system may identify one or more feature points relating to the ROI from the reference image, and determine a motion field of the feature points from the reference motion phase to the target motion phase using a motion prediction model. An input of the motion prediction model may include at least the reference image and the target image. The system may further determine a physiological condition of the ROI based on the motion field.

What is claimed is: 
     
         1 . A system comprising:
 at least one storage device storing a set of instructions for generating a motion prediction model; and   at least one processor configured to communicate with the at least one storage device, wherein when executing the set of instructions, the at least one processor is configured to direct the system to perform operations including:   obtaining at least one training sample, each training sample including a first image and a second image indicative of a physiological motion of a sample region of interest (ROI), the first image corresponding to a first motion phase of the sample ROI, and the second image corresponding to a second motion phase of the sample ROI; and   generating the motion prediction model by training a preliminary model using the at least one training sample according to a machine learning technique, wherein the training of the preliminary model includes one or more iterations, an iteration of the one or more iterations including:   for each of the at least one training sample,
 generating a first motion field from the first image to the second image using the preliminary model in the iteration; 
 generating a predicted second image according to the first motion field; and 
 determining a first difference between the predicted second image and the second image of the training sample; and 
   updating parameter values of the preliminary model to be used in a next iteration based on at least in part on the first difference corresponding to each of the at least one training sample,   wherein the updating parameter values of the preliminary model further comprises:   for each of the at least one training sample,   generating a second motion field from the second image to the first image using the preliminary model; and   determining an opposite motion field of the second motion field.   
     
     
         2 . The system of  claim 1 , wherein the updating parameter values of the preliminary model further comprises:
 determining a value of a loss function based at least in part on the first difference corresponding to each of the at least one training sample; and   updating the parameter values of the preliminary model based on the value of the loss function.   
     
     
         3 . The system of  claim 2 , wherein the at least one processor is further configured to direct the system to perform additional operations including:
 determining a second difference between the opposite motion field and the first motion field of the training sample,   wherein the value of the loss function is determined further based on the second difference corresponding to each training sample.   
     
     
         4 . The system of  claim 3 , wherein the at least one processor is further configured to direct the system to perform additional operations including:
 for each of the at least one training sample,
 generating a predicted first image by warping the second image of the training sample according to the first image of the training sample using the preliminary model; 
 generating a third image by warping the predicted first image according to the second image using the preliminary model; 
 generating a fourth image by warping the predicted second image according to the first image using the preliminary model; and 
 determining a third difference between the third image and the second image and a fourth difference between the fourth image and the first image, 
 wherein the value of the loss function is determined further based on the third difference and the fourth difference corresponding to each training sample. 
   
     
     
         5 . The system of  claim 2 , wherein the preliminary model comprises a generator, and
 for each of the at least one training sample, the generator is configured to predict a first motion field from the first image of the training sample to the second image of the training sample.   
     
     
         6 . The system of  claim 5 , wherein the preliminary model further comprises a transformation layer, and
 for each of the at least one training sample, the transformation layer is configured to warp the first image of the training sample according to the corresponding first motion field to generate the corresponding predicted second image.   
     
     
         7 . The system of  claim 6 , wherein the preliminary model further comprises a discriminator,
 for each of the at least one training sample, the discriminator is configured to generate a discrimination result between the second image of the training sample and the corresponding predicted second image, and   the value of the loss function is determined further based on the discrimination result of each training sample.   
     
     
         8 . The system of  claim 5 , wherein the preliminary model further comprises a second generator, and
 for each training sample, the second generator is configured to predict, based on the first image and the second image of the training sample, a second motion field from the second image of the training sample to the first image of the training sample.   
     
     
         9 . The system of  claim 5 , wherein
 the training the preliminary model includes training the generator, and   the trained generator is designated as the motion prediction model.   
     
     
         10 . The system of  claim 1 , wherein the at least one processor is further configured to direct the system to perform additional operations including:
 obtaining a first annotated image of the sample ROI corresponding a third motion phase and an unannotated image of the sample ROI corresponding a fourth motion phase, the first annotated image including an annotation of a first feature point relating to the ROI;   determining a motion field of the first feature point from the third motion phase to the fourth motion phase by applying the motion prediction model to the first annotated image and the unannotated image; and   generating, based on the annotation of the first feature point and the motion field, a second annotated image of the sample ROI corresponding the fourth motion phase, the second annotated image including an annotation of a second feature point corresponding to the first feature point.   
     
     
         11 . The system of  claim 1 , wherein the at least one processor is further configured to direct the system to perform additional operations including:
 acquiring a reference image of a region of interest (ROI) corresponding to a reference motion phase of the ROI and a target image of the ROI corresponding to a target motion phase of the ROI, the target motion phase being different from the reference motion phase;   identifying one or more feature points relating to the ROI from the reference image;   determining a motion field of the one or more feature points from the reference motion phase to the target motion phase using the motion prediction model, wherein an input of the motion prediction model includes at least the reference image and the target image; and   determining, based on the motion field, a physiological condition of the ROI.   
     
     
         12 . The system of  claim 11 , wherein the ROI includes at least one of a heart, a lung, an abdomen, a chest, a stomach, or of a subject. 
     
     
         13 . The system of  claim 11 , wherein the ROI is a heart, and the one or more feature points relating to the heart in the reference image include a pair of feature points including a first feature point and a second feature point, and to determine a physiological condition of the heart, the at least one processor is further configured to direct the system to perform additional operations including:
 determining, based on the reference image, a first distance between the first feature point and the second feature point in the reference motion phase;   determining, based on the motion vector of the first feature point and the motion vector of the second feature point, a second distance between the first feature point and the second feature point in the target motion phase; and   determining, based on the first distance and the second distance, a strain value relating to the heart.   
     
     
         14 . The system of  claim 1 , wherein the preliminary model is a generative adversarial network (GAN) model. 
     
     
         15 . A method for generating a motion prediction model implemented on a computing device having at least one processor and at least one storage device, the method comprising:
 obtaining at least one training sample, each training sample including a first image and a second image indicative of a physiological motion of a sample region of interest (ROI), the first image corresponding to a first motion phase of the sample ROI, and the second image corresponding to a second motion phase of the sample ROI; and   generating the motion prediction model by training a preliminary model using the at least one training sample according to a machine learning technique, wherein the training of the preliminary model includes one or more iterations, an iteration of the one or more iterations including:   for each of the at least one training sample,
 generating a first motion field from the first image to the second image using the preliminary model in the iteration; 
 generating a predicted second image according to the first motion field; and 
 determining a first difference between the predicted second image and the second image of the training sample; and 
   updating parameter values of the preliminary model to be used in a next iteration based on at least in part on the first difference corresponding to each of the at least one training sample,   wherein the updating parameter values of the preliminary model further comprises:   for each of the at least one training sample,
 generating a second motion field from the second image to the first image using the preliminary model; and 
 determining an opposite motion field of the second motion field. 
   
     
     
         16 . The method of  claim 15 , wherein the updating parameter values of the preliminary model further comprises:
 determining a value of a loss function based on the based on at least in part on the first difference corresponding to each of the at least one training sample; and   updating the parameter values of the preliminary model based on the value of the loss function.   
     
     
         17 . The method of  claim 16 , further comprising:
 determining a second difference between the opposite motion field and the first motion field of the training sample,   wherein the value of the loss function is determined further based on the second difference corresponding to each training sample.   
     
     
         18 . The method of  claim 16 , further comprising:
 for each of the at least one training sample,
 generating a predicted first image by warping the second image of the training sample according to the first image of the training sample using the preliminary model; 
 generating a third image by warping the predicted first image according to the second image using the preliminary model; 
 generating a fourth image by warping the predicted second image according to the first image using the preliminary model; and 
 determining a third difference between the third image and the second image and a fourth difference between the fourth image and the first image, 
 wherein the value of the loss function is determined further based on the third difference and the fourth difference corresponding to each training sample. 
   
     
     
         19 . The method of  claim 15 , wherein the preliminary model comprises a generator, and
 for each of the at least one training sample, the generator is configured to predict a first motion field from the first image of the training sample to the second image of the training sample.   
     
     
         20 . The method of  claim 15 , wherein the preliminary model further comprises a transformation layer, and
 for each of the at least one training sample, the transformation layer is configured to warp the first image of the training sample according to the corresponding first motion field to generate the corresponding predicted second image.